tokenstat
tokenstat

Model catalog

Benchmarks

Arena Elo from free open daily snapshots on GitHub (text, code, vision, agent), plus a sparse Artificial Analysis subset OpenRouter embeds. One score set is cross-linked onto provider clones of the same model. Arena may score the same base model at different reasoning efforts, so those rows stay separate and link to the base model only when it exists in the catalog. The open dumps publish 20 models per Arena board (10 on agent), so a full board is short by design. Empty on a single board is a gap, not a ranking of zero.

19 scored rows on this board · 417 linked into the catalog · as of · Models · AI plans

#ModelScore
1
Anthropic: Claude Fable 5Anthropic · claude-fable-5 · high effort
1506Arena text
2
Anthropic: Claude 4.6 Opus ThinkingAnthropic · claude-opus-4-6 · high effort
1505Arena text
3
Anthropic: Claude 4.7 Opus ThinkingAnthropic · claude-opus-4-7 · high effort
1502Arena text
4
Meta: Muse Spark 1.2Meta · muse-spark-1.2 · xhigh effort
1500Arena text
5
Anthropic: Claude Fable 5.1Anthropic · claude-fable-5.1 · max effort
1498Arena text
6
Anthropic: Claude 4.6 Opus ThinkingAnthropic · claude-opus-4.6
1497Arena text
7
Anthropic: Claude 4.7 Opus ThinkingAnthropic · claude-opus-4-7
1494Arena text
8
Anthropic: Claude Opus 5Anthropic · claude-opus-5 · high effort
1493Arena text
9
Google: Gemini 3.8 FlashGoogle · gemini-3.8-flash · high effort
1493Arena text
10
Meta: Muse Spark 1.1Meta · muse-spark-1.1
1493Arena text
11
Meta: Muse Spark 1.3Meta · muse-spark-1.3 · max effort
1493Arena text
12
Google: Gemini 3.7 FlashGoogle · gemini-3.7-flash · high effort
1490Arena text
13
Anthropic: Claude Opus 5Anthropic · claude-opus-5 · max effort
1487Arena text
14
Google: Gemini 3.1 Pro PreviewGoogle · gemini-3.1-pro-preview
1487Arena text
15
Google: Gemini 3 ProGoogle · gemini-3-pro
1485Arena text
16
Moonshot: Kimi K3Moonshot · kimi-k3 · max effort
1485Arena text
17
Mistral: GLM-5.3Z.ai · glm-5.3 · max effort
1483Arena text
18
OpenAI: GPT-5.6 SolOpenAI · gpt-5.6-sol · xhigh effort
1483Arena text
19
OpenAI: GPT-5.5OpenAI · gpt-5.5 · high effort
1482Arena text

Sources

Arena Elo from Arena AI, via open dumps by oolong-tea-2026. Artificial Analysis indices via OpenRouter (subset). Artificial Analysis for the indices themselves. Plus vendor-reported launch evals collected from announcements, analyst writeups, launch blogs, and community benchmarks. As of .

Turn benchmark scores into working sessions

Run agents in a local Git-enabled workspace, connect to another machine when needed, and track the sessions and tokens behind your work. See what your AI subscriptions covered.

CLI for macOS, Linux, and Windows

Data and list-price estimates may contain mistakes. Not investment advice.